Authors
Wang R; Zheng L; Hawke AL; Carey RE; Breloff SP; Li K; Peng X
Source
Comput Methods Biomech Biomed Eng Imaging Vis 2023 May; 11(3):369-377
Abstract
Residential roofers are often exposed to awkward postures and motions in a prolonged time, which may not only reduce their body stability and increase fall potential, but also increase the risk of musculoskeletal disorders (MSDs). To assess their risks of fatal and musculoskeletal injuries, it is crucial to capture 3D body poses of workers during roofing tasks. In this paper, we proposed a novel two-stage motion estimation approach based on a convolution neural network to estimate residential roofer's body poses using three-view video data. Our approach includes two stages: (1) use of an offline multi-view model to estimate the 3D pose in a single frame; (2) use of a multi-frame model to apply temporal convolutions to refine the multi-view outputs. The performance of the approach was evaluated by comparing our estimation with the gold-standard marker-based 3D human pose during one of the common residential roofing tasks - shingle installation. The evaluation results show that the proposed multi-frame model can effectively improve the accuracy of the coordinate sequence. Moreover, these results prove that the proposed video-based motion estimation approach can efficiently and accurately locate 3D body joints and pave the way for future onsite motion analysis during roofing activities.
Keywords
Biomechanics; Roofers; Construction workers; Motion studies; Body mechanics; Posture; Neural networks; Artificial intelligence; Ergonomics; Musculoskeletal disorders; MSD;
Author Keywords: 3D human pose estimation; video-based motion prediction; residential roofing; deep learning
Contact
Liying Zheng, Health Effects Laboratory Division, National Institute for Occupational Safety and Health, Morgantown, WV, USA
Document Type
Journal Article
Email Address
lzheng2@cdc.gov
Priority Area
Construction
Source Name
Computer Methods in Biomechanics and Biomedical Engineering: Imaging & Visualization
